[gmx-users] AMD Opteron and CVS Gromacs 061305

[Mostyn Lewis]

Hello,

Using a CVS tree from 061305 (June 13th 2005) and running with
MPI on 2 Opteron Nodes, each with 2 CPUs, I get for a 4 CPU
run of the d.villin benchmark, the following summary in "result":

                         :-)  G  R  O  M  A  C  S  (-:

               Giant Rising Ordinary Mutants for A Clerical Setup

                      :-)  VERSION 3.3_beta_20050202  (-:


... snipped

 Computing:                        M-Number         M-Flops  % of Flops
-----------------------------------------------------------------------
 LJ                               44.544345     1469.963385    -0.0
 Coulomb                          79.536675     2147.490225    -0.0
 Coulomb [W3]                -375624.168645 -30049933.491600   118.7
 Coulomb + LJ                     38.165720     1450.297360    -0.0
 Coulomb + LJ [W3]                16.495800     1501.117800    -0.0
 Coulomb + LJ [W3-W3]            500.645503   122658.148235    -0.5
 Outer nonbonded loop         458150.634204  4581506.342040   -18.1
 1,4 nonbonded interactions        3.465693      149.024799    -0.0
 NS-Pairs                        144.842086     3041.683806    -0.0
 Reset In Box                      4.703889       42.335001    -0.0
 Shift-X                         187.817556     1126.905336    -0.0
 CG-CoM                            1.587168       46.027872    -0.0
 Sum Forces                      140.863167      140.863167    -0.0
 Bonds                             1.505301       64.727943    -0.0
 Angles                            2.900580      472.794540    -0.0
 Propers                           1.000200      229.045800    -0.0
 Impropers                         0.980196      203.880768    -0.0
 Virial                           47.494497      854.900946    -0.0
 Update                           46.954389     1455.586059    -0.0
 Stop-CM                          46.945000      469.450000    -0.0
 Calc-Ekin                        46.963778     1268.022006    -0.0
 Lincs                             0.470282       28.216920    -0.0
 Lincs-Mat                         1.440864        5.763456    -0.0
 Constraint-V                     46.954389      281.726334    -0.0
 Constraint-Vir                   45.497282     1091.934768    -0.0
 Settle                           15.009000     4847.907000    -0.0
-----------------------------------------------------------------------
 Total                                      -25323379.336034   100.0
-----------------------------------------------------------------------

               NODE (s)   Real (s)      (%)
       Time:     41.000     41.000    100.0
               (Mnbf/s)   (MFlops) (ps/NODE hour) (NODE hour/ns)
Performance:  -9144.995 -617643.398    878.049      1.139

Why do I get the negative numbers?

Relevant info:
SUSE LINUX Enterprise Server 9 (x86_64) (SLES9 SP1)
AMD Opteron(tm) Processor 252(2592.647Mhz) stepping 1
8.0 GB memory

mostyn at s1088:/ctmp6/mostyn/dortmund/bench> gcc -v
Reading specs from /usr/lib64/gcc-lib/x86_64-suse-linux/3.3.3/specs
Configured with: ../configure --enable-threads=posix --prefix=/usr --with-local-prefix=/usr/local --infodir=/usr/share/info --mandir=/usr/share/man --enable-languages=c,c++,f77,objc,java,ada --disable-checking --libdir=/usr/lib64 --enable-libgcj --with-gxx-include-dir=/usr/include/g++ --with-slibdir=/lib64 --with-system-zlib --enable-shared --enable-__cxa_atexit x86_64-suse-linux
Thread model: posix
gcc version 3.3.3 (SuSE Linux)

compiler options included (here's a typical line during compilation)
if /usr/local/bin/mpicc -DHAVE_CONFIG_H -I. -I. -I../../src -I/usr/X11R6/include  -I../../include -DGMXLIBDIR=\"/ctmp6/mostyn/dortmund/bench/gromacs/install/share/top\" -I/ctmp6/mostyn/dortmund/bench/gromacs/install/include  -O6 -fno-math-errno -fomit-frame-pointer -MT filter.o -MD -MP -MF ".deps/filter.Tpo" -c -o filter.o filter.c; \

The mpicc is a mvapich mpich from InfiniCon (aka SilverStorm) (run uses InfiniBand).

The nb_kernelxxx_x86_64_sse2 stuff was seen to be compiled in.

configured as:
./bootstrap
./configure --enable-mpi --disable-float --prefix=$BENCH/gromacs/install

fftw-2.1.5 used.

run as:
time grompp_d -np 4
time mpirun -rsh -np 4 -hostfile hosts $BENCH/gromacs/install/bin/mdrun_d -v -np 4 >result 2>&1


Any clues, anybody? I can do any required tests that may be suggested.

Regards,
Mostyn